We present observational constraints on the nature of dark energy using the Supernova Legacy Survey three year sample (SNLS3) of Guy et al. (2010) and Conley et al. (2011). We use the 472 SNe Ia in this sample, accounting for recently discovered correlations between SN Ia luminosity and host galaxy properties, and include the effects of all identified systematic uncertainties directly in the cosmological fits. Combining the SNLS3 data with the full WMAP7 power spectrum, the Sloan Digital Sky Survey luminous red galaxy power spectrum, and a prior on the Hubble constant H0 from SHOES, in a flat universe we find omega_m=0.269+/-0.015 and w=-1.061+0.069-0.068 -- a 6.5% measure of the dark energy equation-of-state parameter w. The statistical and systematic uncertainties are approximately equal, with the systematic uncertainties dominated by the photometric calibration of the SN Ia fluxes -- without these calibration effects, systematics contribute only a ~2% error in w. When relaxing the assumption of flatness, we find omega_m=0.271+/-0.015, omega_k=-0.002+/-0.006, and w=-1.069+0.091-0.092. Parameterizing the time evolution of w as w(a)=w_0+w_a(1-a), gives w_0=-0.905+/-0.196, w_a=-0.984+1.094-1.097 in a flat universe. All of our results are consistent with a flat, w=-1 universe. The size of the SNLS3 sample allows various tests to be performed with the SNe segregated according to their light curve and host galaxy properties. We find that the cosmological constraints derived from these different sub-samples are consistent. There is evidence that the coefficient, beta, relating SN Ia luminosity and color, varies with host parameters at >4sigma significance (in addition to the known SN luminosity--host relation); however this has only a small effect on the cosmological results and is currently a sub-dominant systematic.